ABSTRACT Stage III colorectal cancer (CRC) demonstrates substantial variability in tumor biology and clinical outcomes and there is a need to understand prognosis for patients in order to gauge risk benefit for chemotherapy and intensity of chemotherapy administration. These features are not well recapitulated by the current biomarkers in use in the clinic, majority of them are DNA based mutation assays. RNA expression patterns have been described by various investigators and may more fully recapitulate tumor biology. The clinical utility of these findings have been limited by the apparent conflicting subgrouping efforts and lack of a validated gene expression signature as a clinical grade assay applicable on formalin fixed paraffin embedded (FFPE) tissue. In our international collaboration with several academic leaders who have previously published in this field, we have identified a robust consensus subgroup classification based on clustering approaches independent of clinical outcomes. Remarkably, this classification system, termed consensus molecular subtypes (CMS), identified 4 subgroups that provide novel insights into the classification of CRC. One subgroup with mesenchymal, TGF-?, and angiogenic features (CMS4) is associated with a hazard ratio for death of 2.26 (95% CI of 1.41 to 3.61, P=.001), significantly higher than other subgroups, in a multivariate model inclusive of current clinical and pathologic risk factors and genetic signature (Oncotype Dx). We hypothesize that a gene expression signature classifier can be developed and validated for determining the CMS in FFPE tissues, and that this classifier can be implemented to improve prognostication of stage III CRC by classifying them in CMS 4 vs. other subtypes. We have developed a support-vector-machine classifier with very high accuracy for classification based on an Affymetrix array from fresh frozen specimens. We have demonstrated good classification accuracy (>90%) using customized Nanostring codesets on FFPE tumor samples of 85 patients with stage III CRC. We have also demonstrated good technical reproducibility in six of those 85 samples. In this application, we will transfer the assay using the Nanostring Codeset to fresh frozen (FF) and FFPE using a set of paired samples, while maintaining classifier performance. We will then pursue technical and analytic validation of the assay, including precision in repeatability, reproducibility between sample types, inter- lab reproducibility, and impact of RNA quality/quantity. In the UH3 portion of the grant, we will clinically validate the prognostic utility of the gene expression signature assay in single-institution cohort, and then in a completed prospective study of FOLFOX chemotherapy (NRG/NSABPC-08), in a CLIA certified laboratory. Additional data will be used in predicting response to various standard of care therapeutics, which represents a series of future potential applications of the assay. By utilizing an assay developed to classify CRC by its tumor biology, we anticipate development of an enduring tool that will be of greater use than traditional fit-for-purpose tests.